#🧠 Multi-Agent RAG — Document Q&A System

Ask questions about any document and get accurate, cited answers powered by a multi-agent AI pipeline. Upload a PDF, DOCX, or TXT file and instantly chat with it — summarize it, extract key points, or ask anything specific.

---

What it does

• Upload any document — PDF, DOCX, or plain text
• Ask questions in natural language — the system finds the most relevant parts of your document and answers based only on what's actually in it
• Summarize — get a full summary of any document in seconds
• Extract key points — pull out the most important facts and takeaways
• Cited answers — every response tells you which part of the document it came from
• Semantic caching — identical or very similar queries are cached so you don't waste API calls

---

How it works

When you upload a document, it gets split into small chunks and converted into numerical embeddings using a local HuggingFace model. These embeddings are stored in ChromaDB on your machine.

When you ask a question, four agents work together:

Your Question
      ↓
Orchestrator Agent — figures out what you're asking (Q&A, summarize, key points)
      ↓
Retrieval Agent — searches ChromaDB for the most relevant chunks
      ↓
Reasoning Agent — thinks through the retrieved context step by step
      ↓
Generation Agent — writes the final answer with source citations
      ↓
Answer shown in Streamlit UI

---

Tech stack

Layer	Tool	Why
Frontend	Streamlit	Fast, clean UI with no frontend code
LLM	Gemini 1.5 Flash	Free tier — 1M tokens/day
Embeddings	sentence-transformers (HuggingFace)	Runs locally, completely free
Vector DB	ChromaDB	Local, persistent, no setup needed
Framework	LangChain	Agent orchestration and prompt management
Document parsing	PyMuPDF, python-docx	PDF and DOCX support

Total cost to run: $0 — everything except Gemini runs locally. Gemini's free tier gives you 15 requests/minute and 1 million tokens per day.

---

Project structure

multi_agent_rag/
├── app.py                        # Streamlit frontend — run this
├── requirements.txt
├── .env.example                  # Copy this to .env and add your key
│
├── agents/
│   ├── orchestrator.py           # Routes queries to the right agent
│   ├── retrieval_agent.py        # Semantic search over your document
│   ├── reasoning_agent.py        # Chain-of-thought reasoning
│   └── generation_agent.py      # Final answer with citations
│
├── core/
│   ├── document_processor.py    # Ingests and chunks documents
│   ├── embeddings.py            # HuggingFace embedding generation
│   ├── vector_store.py          # ChromaDB interface
│   └── prompt_templates.py      # All LangChain prompts
│
├── utils/
│   ├── text_splitter.py         # Semantic chunking logic
│   └── helpers.py               # Utility functions
│
└── config/
    └── settings.py              # Config loaded from .env

---

Setup

1. Clone the repo

git clone https://github.com/your-username/multi-agent-rag.git
cd multi-agent-rag

2. Install dependencies

pip install -r requirements.txt

3. Get your free Gemini API key

Go to aistudio.google.com/app/apikey, sign in with your Google account, and create a key. It takes about 30 seconds.

4. Set up your environment

cp .env.example .env

Open .env and paste your key:

GEMINI_API_KEY=your_key_here

5. Run the app

streamlit run app.py

Your browser will open automatically at http://localhost:8501.

---

Usage

1. Open the app in your browser
2. Upload a PDF, DOCX, or TXT file using the sidebar
3. Wait a few seconds while the document is processed and indexed
4. Start asking questions in the chat box
5. Use the Summarize or Key Points buttons for instant one-click analysis
6. Expand the Sources section under any answer to see exactly which part of the document was used

---

Example queries

• "What is the main argument of this paper?"
• "List all the dates and deadlines mentioned."
• "What does the author recommend in the conclusion?"
• "Summarize section 3."
• "Are there any risks or limitations mentioned?"

---

Environment variables

Variable	Description
GEMINI_API_KEY	Your Google Gemini API key (required)
CHUNK_SIZE	Token size per document chunk (default: 500)
CHUNK_OVERLAP	Overlap between chunks (default: 50)
TOP_K_RESULTS	Number of chunks retrieved per query (default: 5)
CACHE_SIMILARITY_THRESHOLD	Cosine similarity above which a cached answer is reused (default: 0.95)

---

Known limitations

• Works best with text-heavy documents — scanned image PDFs without OCR won't be indexed properly
• Very large documents (100+ pages) may take 30–60 seconds to process on first upload
• Gemini free tier has a rate limit of 15 requests per minute — if you hit it, just wait a moment and retry

---

Built with

• LangChain
• Google Gemini
• ChromaDB
• Streamlit
• sentence-transformers

---

License

MIT — free to use, modify, and distribute.